AGI
Artificial General Intelligence (AGI) is a hypothetical stage in the development of machine learning (ML) in which an artificial intelligence (AI) system can match or exceed the cognitive abilities of human beings across any task.
Understanding AGI: When AI Becomes Truly General
Quick Explanation
Publicly available AI: like ChatGPT, Claude, Mistral, and others, called “Narrow AI”
Like a hammer that is great for one specific task. Of course, you can use a hammer for different tasks, like making music, as a weapon, or for sports activities. But these uses are more about the user’s creativity than the tool’s versatility.
Artificial General Intelligence (AGI): no example yet
Like a multifunctional tool that can create anything without human control. When necessary, it can learn new skills or new ways to solve problems. It can adapt to new and changing environments. It’s like a well-trained and resourceful human being.
Extended explanation
In the age of AI, most of us feel we’ve already reached an impossible level of artificial intelligence. AI tools can write lengthy texts, create images and films, generate programming code, and chat with us like intelligent, thoughtful partners.
However, officially, the publicly available AI tools belong to the “Narrow AI” category. This means these tools are specialized, they can produce high-quality output (text, pictures, code) in only one field. If an AI tool excels at writing, it doesn’t necessarily mean it can write proper code or solve complex mathematical problems.
These narrow AI tools also cannot autonomously learn new tasks or expand their problem-solving capabilities beyond their training. A great image generator couldn’t learn to play piano or write world-class novels.
Of course, we all know people who lack such capabilities too, but officially, human intelligence is considered broader and more general than today’s most advanced AI.
What Must AI Achieve to Become AGI?
So what levels must today’s AI reach or exceed to become Artificial General Intelligence (AGI)? There are several different approaches to defining AGI:
The Turing Test: Machines that can convincingly act like humans in conversation.
Strong AI: Systems that possess consciousness and self-awareness.
Human-level performance: AI that matches humans on cognitive tasks across multiple domains.
Learning ability: Systems that can autonomously learn entirely new tasks.
Economic value: AI that can perform economically valuable work across various fields.
Flexible capabilities: Perhaps the most practical definition comes from Apple co-founder Steve Wozniak, who asked: “Could a computer make a cup of coffee?” This seemingly simple task is actually quite complex: one must be able to walk, understand what kitchens are, recognize coffee machines, and interact with drawers and cabinets. Humans draw upon a lifetime of experience just to brew a cup of coffee.
“Artificial Capable Intelligence” (ACI): Some propose a “Modern Turing Test” where an AI system receives $100,000 in seed capital and must grow it into $1 million, demonstrating the ability to accomplish complex, open-ended, multistep tasks in the real world.
The Promise of AGI
AGI could revolutionize fields like robotics, healthcare, education, and scientific research by enabling AI systems to tackle complex challenges that currently require human-level reasoning and adaptability.
When Will We Reach This Level?
Short answer: We don’t know.
Longer answer: Predictions about AI’s future are highly uncertain, but nearly all experts agree AGI will be possible by the end of this century, with some estimating it might happen much sooner.
The debate is already underway. Some researchers, such as Blase Agüera y Arcas and Peter Norvig, argue that advanced large language models like Meta’s Llama, OpenAI’s GPT, and Anthropic’s Claude have already achieved AGI.
However, Yann LeCun, Meta’s chief AI scientist, disagrees. He argues that current LLMs lack AGI because they don’t have common sense, they can’t think before they act, can’t perform actions in the real world, can’t learn through embodied experience, and lack persistent memory and capacity for hierarchical planning. LeCun and Jacob Browning have argued that “a system trained on language alone will never approximate human intelligence, even if trained from now until the heat death of the universe.”
A Historical Perspective
Here’s a reminder about the unpredictability of technological breakthroughs: Wilbur Wright once proclaimed, “I confess that in 1901, I said to my brother Orville that men would not fly for 50 years. Two years later, we were making flights.”
The future of AGI may surprise us all, perhaps sooner than we think, or in ways we never expected.